Matter Made Labs
Accelerate materials
science
Get tangible business results in
R&D and Operations with AI
Schedule Strategy CallCreating new materials is hard. MMLabs can help.
Manufacturing and materials innovations often requires many experiments.
Many parameters to tune
Often projects see between 8-20 process parameters, and 5-20 different ingredients to combine. This is high dimensional space that is hard for humans to keep track of.
Multiple constraints to balance:
Typical projects have >5 different targets. Yield strength and toughness and cost and weight.
Small datasets
Many projects start with <20 data points. This is insufficient for most AI algorithms.
Messy data
Most lab scale or prototype scale tests are not in a nice table. They are in different formats, in various .csv files or maybe just in lab notebooks.
Case Studies
€10 million/year saved in chemical processing plant
Challenge: A chemical manufacturer needed to fine tune the settings of the pipeline for each new material batch, with optimization taking 3 days while new batches arrived every 4-5 days.
Key Insights:
- An excellent simulation existed but was underutilized due to 30-120 minute run times
- Manual safety checks added a half-day delay
- Process experts were operating on intuition rather than systematic optimization
Implementation:
Developed an AI-powered simulation replica that ran in <0.1 seconds with 98% accuracy
Systematized 100+ safety checks into an automated verification system that experts could quickly approve
Created optimization algorithms that consistently outperformed manual tuning
Results:
€10M Annual Savings per plant through optimized production
99% Reduction in parameter optimization time (3 days → 20 minutes)
Zero Safety Incidents during the transition to AI-assisted parameter settings
Materials breakthrough in 90 days
Challenge: A specialty glass manufacturer had tested 340 formulations over 18 months with minimal progress. Their best formula achieved only 5 out of 12 mandatory properties.
Key Insights:
- Risk aversion had led to incremental experimentation that would never reach the required properties
- 20% of experiments were duplicates or near-duplicates.
- Scientists spent 10+ hours weekly on experiment design rather than execution.
Implementation:
- Created a novel visualization tool revealing that alternative substrate paths had higher potential
- Implemented active learning algorithms to systematically explore the parameter space
- Automated experiment tracking to eliminate duplication
Results:
Stage-approved formula in 90 days: Achieved all 12 mandatory properties
80% Reduction in experiment design time
Transformed R&D Strategy toward more ambitious material exploration
2 year time saving in development of new polymer
Challenge: A multinational polymer manufacturer needed to develop a fire-resistant polymer blend, with 30,000 potential candidates but capacity for only 3 experiments weekly.
Key Insights: While direct property prediction wasn't possible with available data, auxiliary correlations could be leveragedThe lead scientist spent 10+ hours weekly formatting data and running basic analysisPrevious approach focused on a promising but fundamentally limited chemical family
Implementation:
Developed multi-property prediction models using auxiliary data, improving accuracy from 10% to 85-95%
Automated data formatting and analysis, reducing scientist workload by 95% (5 hours per week -> 10 minutes)
Created visualization tools to compare candidate pathways systematically
Results:
2-Year Development Time Savings according to lead scientist
4X Higher Predictive Accuracy for critical material properties
25% More Experiments Run through increased scientist productivity
€4 Million in additional throughput
Challenge: A multi-national food and beverage manufacturer needed to optimize their aging manufacturing line. If buffer lines were not used effectively machines would get starved or blocked.
Key Insights:
- The optimization algorithm was local; only looking at the nearest two machines.
- There is significant data in the error messages; knowing why a machine stalls lets you know whether its likely that it will run in 1 min or in 10 minutes.
Implementation:
- Created a simulation based on the data of their manufacturing plant
- Developed AI algorithms to steer the simulation-based manufacturing line
- Deployed the AI algorithm on the factory line after extensive testing
Results:
€4 Million additional throughput per year from one production line.
Solution
Analytics tools have come a very long way. Leverage them to unblock your most important R&D work and accelerate your time to market.
- Run only the most valuable experiments. Bayesian Optimization is a mature technology that is proven to accelerate materials development and discovery. Identify the top most valuable experiments to run before committing days or weeks of lab time.
- Find relevant information from internal/external reports. If you have data in internal reports that are hard to access or search through, or externally published papers but there are too many to go through, there are now robust tools that help your researchers find what they need faster.
- Use data analysis tools: LLMs are very good at doing data analysis. Give everyone in your team a pocket data scientist. Let them enable your materials teams by having LLM create analysis reports and figures. The LLM just generates code that executes on your data, so there is no risk of hallucinations.
About me
Triple master's degrees from MIT in Computer Science, Advanced Manufacturing, and in Technology Policy
Real implementations: Deployed dozens of AI solutions for Fortune 500 and FT Europe 500 companies. Delivered projects that automates 1000s of hours of work per year, €10 million/year in savings, €4 million/year in additional throughput, and more.
Cutting edge AI experience: Worked at the trailblazing first Citrine Informatics (the first materials informatics company), at Microsoft Copilot developing the first AI data scientists, and Microsoft Autonomous Systems - using AI to streamline manufacturing processing and operations.
Not Ready to Talk Yet?
Share your ideas and questions
and I'll provide initial thoughts
